Travellers in the UAE plan trips with AI. A hotel playbook before booking moves into the answer

Why the UAE is a market to prepare first

Travellers in the UAE already plan trips with AI. Marriott Bonvoy surveyed 4,017 adults in the UAE and Saudi Arabia, and TradeArabia reported the results in November 2025. In that survey (fieldwork dates not stated), 77% of UAE respondents said they had used AI to plan or research a holiday.¹

The wider population is just as quick to adopt. The National reported in May 2026 that Microsoft's AI diffusion ranking put the UAE first, with a diffusion score of 70.1%, a measure of how widely working-age people use AI tools (data period not stated in the article).² Singapore, Norway, Ireland and France followed. A traveller who uses AI at work is likely to use it to plan a weekend away.

Read the survey with its source in mind. Marriott sells hotel stays and has a stake in travellers booking online, and the figures come through a trade-press report rather than the full report. Still, the direction matches every other signal from the market. Our view: for hotels and operators selling to UAE residents and to visitors arriving there, AI answers are already part of the shortlist, and the question is whether yours are accurate.

This playbook covers what travellers ask and why travel answers cite more sources than most. It then looks at how Google's own cards and its new booking flow change the path, and what to fix in English and Arabic over the next ninety days.

Travellers who have used AI to plan or research a holiday

UAE respondents¹77%

UAE and Saudi Arabia combined¹72%

Saudi Arabia respondents¹68%

What do travellers in the UAE ask AI?

They ask comparison questions with constraints, in English and in Arabic. The traveller rarely asks for a hotel by name. They describe a trip and let the engine build the shortlist.

The same survey reported that most respondents were comfortable with the idea of AI booking accommodation for them, with fieldwork dates again not stated.¹ That matters for how prompts are phrased. A traveller who expects to book in the chat asks for availability, price and terms, not only for ideas.

Three kinds of traveller ask these questions. Residents plan staycations and short regional trips. Visitors from the rest of the Gulf plan family holidays. Long-haul visitors from Europe, India and the United States compare the UAE with other winter-sun destinations. Each group writes prompts differently, so a prompt set built only from the hotel's marketing language will miss most of them.

English and Arabic answers are different samples

An answer in Arabic is not a translation of the answer in English. Engines retrieve different sources and weigh them differently by language. A July 2026 preprint by Żatuchin studied why AI answers about brands vary. It found that the language of the query was the largest systematic source of variation it measured, far larger than the brand itself.³ It is one study of brands in another region, so treat it as a reason to measure, not a fixed result.

Our view: a hotel selling to Arabic-speaking guests needs a separate Arabic prompt set, run and reported on its own. Averaging English and Arabic answers into one score hides the language where you are weakest.

Travel answers cite more than most

Travel is one of the sectors where AI answers link out most, which makes it unusually open to a hotel's own pages. In Similarweb's US data for June 2025 to May 2026, about 23% of ChatGPT prompts in travel and hospitality carried a citation, against under 4% in professional services. Across all prompts the share had reached about 6.8% by May 2026.⁴

The figures are US, not UAE, and no comparable UAE breakdown has been published that we know of. The mechanism is unlikely to differ much. Travel questions depend on prices, dates and availability that change, so engines look them up rather than answer from memory.

A citation is a door. When an engine links a page to support "free cancellation until 48 hours before arrival", the traveller can click through to that page and book direct. If the page that carries the fact belongs to an agency, the door opens onto the agency. Our note on citation rates by sector explains the Similarweb data in more detail.

AI answers and travel, 2026

Where travellers meet hotels in AI answers

About 23%of US ChatGPT travel prompts carried a citation, Jun 2025 to May 2026Similarweb ⁴

8.4xgrowth in google.com's citation share in AI Mode, Apr to Jun 2026Profound ⁷

1B+monthly AI Mode users, by Google's own count, May 2026Google ⁶

Google's own cards now sit between traveller and hotel

In Google's AI Mode, the page cited for a hotel is increasingly Google's own. In Profound's data for 15 April to 30 June 2026, google.com's citation share in AI Mode grew 8.4 times, making it the second most-cited domain.⁷ Profound traced the growth to Business Profile and product knowledge panel cards. In an analysis of more than 32 million Business Profile citations, it listed hospitality and travel first among the most disrupted sectors.

AI Mode is not a US curiosity. Google said in October 2025 that AI Mode was available in more than 200 countries and territories, and in May 2026 it said AI Mode had passed one billion monthly users.⁵⁶ Both are company statements, but they set the scale.

For a hotel, this makes the Business Profile a second homepage. The card carries category, photos, amenities, reviews and replies. If the card is thin, the answer is thin. If reviews mention a renovation that finished last year and nobody replied, the answer may still warn travellers about building work.

Our view: most hotels in the region maintain the Business Profile less carefully than their website, because it used to matter only for map searches. That order should now be reversed for the facts travellers compare.

Booking is moving into the answer

Google has started letting travellers book hotels without leaving AI Mode, in the US first. According to Skift, from 27 August 2026 US travellers can compare rooms, review cancellation terms and pay with Google Pay inside AI Mode.⁸ The bookings run through partners including Booking.com, Expedia, Hilton, IHG and Marriott, and the hotel or agency stays merchant of record.

Google has not said when the feature will reach other countries. Its earlier plans point the same way: at I/O in May 2026 it described agentic booking for local experiences and services, starting in the US that summer.⁶ Operators selling desert tours, boat trips and attraction tickets should expect the same pattern to reach them.

When booking arrives in the UAE, the hotels in the flow will be those connected through a partner. Independent hotels and smaller operators will mostly be bookable through agencies, at agency commission, or described in the answer with a link to their own site. Our note on the US launch covers what that means for independents.

Our view: prepare for the booking flow by fixing the facts it reads, not by waiting for an integration you cannot control. A booking agent compares rooms on price, terms and fit. Whichever page states those most clearly, yours or an agency's, frames the comparison.

The playbook: what to fix first

Fix the facts that an engine compares and a traveller checks, on pages you own, in both languages. Six areas cover most of the gap.

Rates and cancellation terms in plain text

Publish a "from" rate per room type and season as text on your own site, and keep it current. Write the cancellation policy as one clear paragraph per rate type. Booking engines often show these only inside a booking widget, which some crawlers cannot read. A sentence in the page body can be read and cited.

Room and property facts

List beds, maximum occupancy, size, view, accessibility features and what the rate includes. Mark up the property with structured data using the schema.org Hotel type and its related room and offer types, as described on schema.org.⁹ Structured data helps machines parse facts; the visible text is what answers quote.

Proper Arabic pages

Publish real Arabic pages for the main property, room and policy content, written or reviewed by a fluent editor. Lay them out right to left, with language tags linking each page to its English version. Machine translation of a booking widget is not enough. If the Arabic page says something different from the English one, engines will repeat the difference.

Business Profile

Set the correct category, add current photos of rooms and facilities, fill every amenity attribute and reply to recent reviews, including those written in Arabic. Check the profile monthly; guests and the platform can change details without telling you.

Listings that agree

Make sure every agency listing carries the same room names, rates logic, terms and amenities as your own site. When sources disagree, engines pick one, and it may be the stale one.

Third-party sources engines cite

Find which travel guides, review sites and publishers the engines cite for your prompts, and make sure they describe you correctly. Our checklist for making travel sites machine-readable covers crawler access and page structure in more depth.

Our recommended order for a single property; groups can run the rows in parallel across hotels.

Two ways to handle Arabic prompts

One blended score

• English and Arabic answers averaged together
• Translated widget text, no Arabic policy pages
• Weak language hidden by the strong one

Two samples, two plans

• Separate prompt sets, run and reported per language
• Edited Arabic pages for rooms, rates and terms
• Fixes prioritised by where each language is weakest

How to measure it

Measure the UAE as its own market, with separate prompt sets for English and Arabic and for the traveller groups you sell to. A single global score will not tell a hotel which language or which guest type is the problem.

Our audits run each prompt on three separate days across six engines: ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI. We use official APIs where an engine offers one and a licensed AI-answer data provider otherwise. For a hotel, the useful outputs are share of answer per language, the sources each engine cites, and a register of wrong facts such as old rates or missing amenities.

Watch the wrong-facts register closely. Hotels change faster than most businesses: rates move by season, restaurants close for refurbishment, shuttle services start and stop. An engine that learned last year's facts will repeat them until a fresher, clearer source replaces them. Each wrong fact in the register should map to a page you can fix and a date you fixed it.

Separate the guest types in reporting too. A resort may be strong in English answers for long-haul family holidays and invisible in Arabic answers for Gulf weekend breaks. Those need different pages and different outside sources, and a blended score would show neither problem.

Budget follows the number of markets and languages. Our guide to budgeting AI visibility work across countries shows how to decide between one market and several. After the baseline, run the prompt set weekly and report monthly, as our note on re-run cadence recommends. Add an extra run when Google's booking flow or a new model reaches the region.

A ninety-day plan

Ninety days is enough to fix the facts and measure the change, if one person owns the plan. The order below puts measurement first so that every fix has a before and after.

Ninety days for a hotel selling to UAE travellers

1. Weeks 1 to 2

   Build English and Arabic prompt sets by guest type; run a baseline across engines on separate days.

2. Weeks 3 to 4

   Fix rates, cancellation terms and room facts as text; update the Business Profile.

3. Weeks 5 to 8

   Publish edited Arabic pages; align every agency listing with your own site.

4. Weeks 9 to 10

   Correct the third-party pages engines cite for your prompts.

5. Weeks 11 to 13

   Re-run the frozen prompt sets and compare with the baseline, language by language.

Tour operators can follow the same plan with one change: replace room facts with experience facts. Duration, pick-up, inclusions, age limits and cancellation terms are what an engine compares when a traveller asks for "a half-day desert trip for a family with young children".

What to watch next

Three things will tell you how fast to move. First, any announcement that Google's AI Mode booking is reaching new countries. Second, changes in which domains engines cite for your Arabic prompts, which can shift as more Arabic content is published. Third, your own direct-booking share from AI referrals, which shows whether the fixes are sending guests to you or to agencies.

None of this requires a new booking platform. It requires accurate facts, written as text, in the languages your guests use, on pages engines can read. Hotels that do that work before booking moves into the answer will be described correctly when it does.